def _block_dim_adaptive(block_dim_adaptive):
"""Enable/Disable backends set block_dim adaptively.
"""
if get_runtime().prog.config.arch != cpu:
_logging.warn('Adaptive block_dim is supported on CPU backend only')
else:
get_runtime().prog.config.cpu_block_dim_adaptive = block_dim_adaptive
def __call__(self, *args, **kwargs):
if self.is_grad and impl.current_cfg().opt_level == 0:
_logging.warn(
"""opt_level = 1 is enforced to enable gradient computation."""
)
impl.current_cfg().opt_level = 1
assert len(kwargs) == 0, 'kwargs not supported for Taichi kernels'
key = self.ensure_compiled(*args)
return self.compiled_functions[key](*args)
def __call__(self, *args, **kwargs):
args = _process_args(self, args, kwargs)
if self.is_grad and impl.current_cfg().opt_level == 0:
_logging.warn(
"""opt_level = 1 is enforced to enable gradient computation."""
)
impl.current_cfg().opt_level = 1
key = self.ensure_compiled(*args)
return self.compiled_functions[key](*args)
def adaptive_arch_select(arch, enable_fallback, use_gles):
if arch is None:
return cpu
if not isinstance(arch, (list, tuple)):
arch = [arch]
for a in arch:
if is_arch_supported(a, use_gles):
return a
if not enable_fallback:
raise RuntimeError(f'Arch={arch} is not supported')
_logging.warn(f'Arch={arch} is not supported, falling back to CPU')
return cpu
def block_local(*args):
"""Hints Taichi to cache the fields and to enable the BLS optimization.
Please visit https://docs.taichi-lang.org/docs/performance
for how BLS is used.
Args:
*args (List[Field]): A list of sparse Taichi fields.
"""
if impl.current_cfg().opt_level == 0:
_logging.warn("""opt_level = 1 is enforced to enable bls analysis.""")
impl.current_cfg().opt_level = 1
for a in args:
for v in a._get_field_members():
get_runtime().prog.current_ast_builder().insert_snode_access_flag(
_ti_core.SNodeAccessFlag.block_local, v.ptr)
def _clamp_unsigned_to_range(npty, val):
# npty: np.int32 or np.int64
iif = np.iinfo(npty)
if iif.min <= val <= iif.max:
return val
cap = (1 << iif.bits)
if not 0 <= val < cap:
# We let pybind11 fail intentionally, because this isn't the case we want
# to deal with: |val| does't fall into the valid range of either
# the signed or the unsigned type.
return val
new_val = val - cap
warn(
f'Constant {val} has exceeded the range of {iif.bits} int, clamped to {new_val}'
)
return new_val
def __call__(self, *args, **kwargs):
args = _process_args(self, args, kwargs)
# Transform the primal kernel to forward mode grad kernel
# then recover to primal when exiting the forward mode manager
if self.runtime.fwd_mode_manager:
# TODO: if we would like to compute 2nd-order derivatives by forward-on-reverse in a nested context manager fashion,
# i.e., a `Tape` nested in the `FwdMode`, we can transform the kernels with `mode_original == AutodiffMode.REVERSE` only,
# to avoid duplicate computation for 1st-order derivatives
mode_original = self.autodiff_mode
self.autodiff_mode = AutodiffMode.FORWARD
self.runtime.fwd_mode_manager.insert(self, mode_original)
if self.autodiff_mode != AutodiffMode.NONE and impl.current_cfg(
).opt_level == 0:
_logging.warn(
"""opt_level = 1 is enforced to enable gradient computation."""
)
impl.current_cfg().opt_level = 1
key = self.ensure_compiled(*args)
return self.runtime.compiled_functions[key](*args)
import functools
import os
import sys
from taichi._logging import info, warn
from taichi.core.util import ti_core as _ti_core
try:
import sourceinspect as oinspect # pylint: disable=unused-import
except ImportError:
warn('`sourceinspect` not installed!')
warn(
'Without this package Taichi may not function well in Python IDLE interactive shell, '
'Blender scripting module and Python native shell.')
warn('Please run `python3 -m pip install sourceinspect` to install.')
import inspect as oinspect # pylint: disable=unused-import
pybuf_enabled = False
_env_enable_pybuf = os.environ.get('TI_ENABLE_PYBUF', '1')
if not _env_enable_pybuf or int(_env_enable_pybuf):
# When using in Jupyter / IDLE, the sys.stdout will be their wrapped ones.
# While sys.__stdout__ should always be the raw console stdout.
pybuf_enabled = sys.stdout is not sys.__stdout__
_ti_core.toggle_python_print_buffer(pybuf_enabled)
def _shell_pop_print(old_call):
if not pybuf_enabled:
# zero-overhead!
return old_call
def init(arch=None,
default_fp=None,
default_ip=None,
_test_mode=False,
enable_fallback=True,
**kwargs):
"""Initializes the Taichi runtime.
This should always be the entry point of your Taichi program. Most
importantly, it sets the backend used throughout the program.
Args:
arch: Backend to use. This is usually :const:`~taichi.lang.cpu` or :const:`~taichi.lang.gpu`.
default_fp (Optional[type]): Default floating-point type.
default_ip (Optional[type]): Default integral type.
**kwargs: Taichi provides highly customizable compilation through
``kwargs``, which allows for fine grained control of Taichi compiler
behavior. Below we list some of the most frequently used ones. For a
complete list, please check out
https://github.com/taichi-dev/taichi/blob/master/taichi/program/compile_config.h.
* ``cpu_max_num_threads`` (int): Sets the number of threads used by the CPU thread pool.
* ``debug`` (bool): Enables the debug mode, under which Taichi does a few more things like boundary checks.
* ``print_ir`` (bool): Prints the CHI IR of the Taichi kernels.
* ``packed`` (bool): Enables the packed memory layout. See https://docs.taichi.graphics/lang/articles/advanced/layout.
"""
# Check version for users every 7 days if not disabled by users.
skip = os.environ.get("TI_SKIP_VERSION_CHECK")
if skip != 'ON':
# We don't join this thread because we do not wish to block users.
check_version_thread = threading.Thread(target=try_check_version,
daemon=True)
check_version_thread.start()
# Make a deepcopy in case these args reference to items from ti.cfg, which are
# actually references. If no copy is made and the args are indeed references,
# ti.reset() could override the args to their default values.
default_fp = _deepcopy(default_fp)
default_ip = _deepcopy(default_ip)
kwargs = _deepcopy(kwargs)
reset()
spec_cfg = _SpecialConfig()
env_comp = _EnvironmentConfigurator(kwargs, cfg)
env_spec = _EnvironmentConfigurator(kwargs, spec_cfg)
# configure default_fp/ip:
# TODO: move these stuff to _SpecialConfig too:
env_default_fp = os.environ.get("TI_DEFAULT_FP")
if env_default_fp:
if default_fp is not None:
_ti_core.warn(
f'ti.init argument "default_fp" overridden by environment variable TI_DEFAULT_FP={env_default_fp}'
)
if env_default_fp == '32':
default_fp = f32
elif env_default_fp == '64':
default_fp = f64
elif env_default_fp is not None:
raise ValueError(
f'Invalid TI_DEFAULT_FP={env_default_fp}, should be 32 or 64')
env_default_ip = os.environ.get("TI_DEFAULT_IP")
if env_default_ip:
if default_ip is not None:
_ti_core.warn(
f'ti.init argument "default_ip" overridden by environment variable TI_DEFAULT_IP={env_default_ip}'
)
if env_default_ip == '32':
default_ip = i32
elif env_default_ip == '64':
default_ip = i64
elif env_default_ip is not None:
raise ValueError(
f'Invalid TI_DEFAULT_IP={env_default_ip}, should be 32 or 64')
if default_fp is not None:
impl.get_runtime().set_default_fp(default_fp)
if default_ip is not None:
impl.get_runtime().set_default_ip(default_ip)
# submodule configurations (spec_cfg):
env_spec.add('log_level', str)
env_spec.add('gdb_trigger')
env_spec.add('experimental_real_function')
env_spec.add('short_circuit_operators')
env_spec.add('ndarray_use_torch')
# compiler configurations (ti.cfg):
for key in dir(cfg):
if key in ['arch', 'default_fp', 'default_ip']:
continue
_cast = type(getattr(cfg, key))
if _cast is bool:
_cast = None
env_comp.add(key, _cast)
unexpected_keys = kwargs.keys()
if len(unexpected_keys):
raise KeyError(
f'Unrecognized keyword argument(s) for ti.init: {", ".join(unexpected_keys)}'
)
# dispatch configurations that are not in ti.cfg:
if not _test_mode:
set_gdb_trigger(spec_cfg.gdb_trigger)
impl.get_runtime().experimental_real_function = \
spec_cfg.experimental_real_function
impl.get_runtime().short_circuit_operators = \
spec_cfg.short_circuit_operators
impl.get_runtime().ndarray_use_torch = \
spec_cfg.ndarray_use_torch
_logging.set_logging_level(spec_cfg.log_level.lower())
# select arch (backend):
env_arch = os.environ.get('TI_ARCH')
if env_arch is not None:
_logging.info(f'Following TI_ARCH setting up for arch={env_arch}')
arch = _ti_core.arch_from_name(env_arch)
cfg.arch = adaptive_arch_select(arch, enable_fallback, cfg.use_gles)
if cfg.arch == cc:
_ti_core.set_tmp_dir(locale_encode(prepare_sandbox()))
print(f'[Taichi] Starting on arch={_ti_core.arch_name(cfg.arch)}')
# Torch based ndarray on opengl backend allocates memory on host instead of opengl backend.
# So it won't work.
if cfg.arch == opengl and spec_cfg.ndarray_use_torch:
_logging.warn(
'Opengl backend doesn\'t support torch based ndarray. Setting ndarray_use_torch to False.'
)
impl.get_runtime().ndarray_use_torch = False
if _test_mode:
return spec_cfg
get_default_kernel_profiler().set_kernel_profiler_mode(cfg.kernel_profiler)
# create a new program:
impl.get_runtime().create_program()
_logging.trace('Materializing runtime...')
impl.get_runtime().prog.materialize_runtime()
impl._root_fb = FieldsBuilder()
if not os.environ.get("TI_DISABLE_SIGNAL_HANDLERS", False):
impl.get_runtime()._register_signal_handlers()
return None
